How AI Is Unlocking the Secrets of Nature and the Universe | Demis Hassabis | TED

翻译人员: Yip Yan Yeung 校对人员: Lening Xu

克里斯·安德森（Chirs Anderson）： 戴密斯，欢迎来到这里。

Chris Anderson: Demis, so good to have you here.

Demis Hassabis: It's fantastic to be here, thanks, Chris.

戴密斯·哈萨比斯 （Demis Hassabis）：

来到这里真是太棒了， 谢谢，克里斯。

Now, you told Time Magazine,

你告诉《时代》杂志，

"I want to understand the big questions,

“我想了解那些重大问题，

the really big ones that you normally go into philosophy or physics

那些如果你有兴趣， 会涉及哲学或物理的大问题。

if you're interested in them.

I thought building AI

我认为构建 AI

would be the fastest route to answer some of those questions."

将是回答其中一些问题的最快途径。”

Why did you think that?

你为什么这么想？

DH: (Laughs)

DH：（笑）

Well, I guess when I was a kid,

我想是因为在我小时候，

my favorite subject was physics,

我最喜欢的科目是物理学，

and I was interested in all the big questions,

我对所有重大问题、

fundamental nature of reality,

现实的基本本质、

what is consciousness,

什么是意识，

you know, all the big ones.

所有重大问题都很感兴趣。

And usually you go into physics, if you're interested in that.

通常如果你喜欢物理， 你就会对这些问题感兴趣。

But I read a lot of the great physicists,

但我了解了很多伟大的物理学家，

some of my all-time scientific heroes like Feynman and so on.

有些是我这一生的科学英雄， 比如费曼（Feynman）等等。

And I realized, in the last, sort of 20, 30 years,

我意识到，在过去的二三十年中，

we haven't made much progress

我们在理解一些基本定律方面 并没有取得太大的进展。

in understanding some of these fundamental laws.

So I thought, why not build the ultimate tool to help us,

所以我想，为什么不开发 一个终极工具来帮助我们，

which is artificial intelligence.

那就是人工智能。

And at the same time,

同时，

we could also maybe better understand ourselves

我们也可以借此 更好地了解自己和大脑。

and the brain better, by doing that too.

So not only was it incredible tool,

它不仅是个神奇的工具，

it was also useful for some of the big questions itself.

而且对一些重大问题本身也很有用。

CA: Super interesting.

CA：非常有意思。

So obviously AI can do so many things,

显然 AI 可以做很多事情，

but I think for this conversation,

但我想在这次对话中，

I'd love to focus in on this theme of what it might do

我们来重点讨论它可以做些什么

to unlock the really big questions, the giant scientific breakthroughs,

来解开真正的重大问题、 巨大的科学突破，

because it's been such a theme driving you and your company.

因为它就是推动你 和你的公司前进的那个主题。

DH: So I mean, one of the big things AI can do,

DH：我想说，我一直在思考 AI 能做的的一件大事是

and I've always thought about,

is we're getting, you know, even back 20, 30 years ago,

即使是二三十年前，

the beginning of the internet era and computer era,

互联网时代和计算机时代的开始，

the amount of data that was being produced

产生的数据量及科学数据，

and also scientific data,

just too much for the human mind to comprehend in many cases.

很多时候，人脑是无法理解的。

And I think one of the uses of AI is to find patterns and insights

我认为 AI 的用途之一 是在大量数据中

in huge amounts of data and then surface that

找到规律和见解，然后将其呈现

to the human scientists to make sense of

给人类科学家，让他们理解

and make new hypotheses and conjectures.

并做出新的假设和猜想。

So it seems to me very compatible with the scientific method.

在我看来，它与科学方法非常兼容。

CA: Right.

CA：对。

But game play has played a huge role in your own journey

但是“玩游戏”在你搞明白这些的途中 扮演了重要的角色。

in figuring this thing out.

Who is this young lad on the left there?

左边的这位小伙子是谁？

Who is that?

那是谁？

DH: So that was me, I think I must have been about around nine years old.

DH：是我，我猜大概 9 岁。

I'm captaining the England Under 11 team,

我是英格兰 11 岁以下 国际象棋队的队长

and we're playing in a Four Nations tournament,

我们正在参加四国锦标赛，

that's why we're all in red.

这就是为什么 我们都穿着红色的衣服。

I think we're playing France, Scotland and Wales, I think it was.

我想我们在打法国、苏格兰 和威尔士，没错。

CA: That is so weird, because that happened to me too.

CA：这太奇怪了， 因为这也发生在了我身上。

In my dreams.

在我的梦里。

(Laughter)

（笑声）

And it wasn't just chess,

不只是国际象棋，

you loved all kinds of games.

你喜欢各种各样的游戏。

DH: I loved all kinds of games, yeah.

DH：是的，我喜欢各种各样的游戏。

CA: And when you launched DeepMind,

CA：当你推出 DeepMind 时，

pretty quickly, you started having it tackle game play.

你很快就开始让它解决游戏问题。

Why?

为什么？

DH: Well, look, I mean, games actually got me into AI in the first place

DH：首先，是游戏把我带进了 AI，

because while we were doing things like,

因为我们做了一些事情，

we used to go on training camps with the England team and so on.

我们经常和英格兰队 一起参加训练营等等。

And actually back then,

其实当时，

I guess it was in the mid '80s,

我猜是在 80 年代中期，

we would use the very early chess computers,

如果你还有印象，

if you remember them, to train against,

我们会使用早期的国际象棋计算机 来互相训练和对战。

as well as playing against each other.

And they were big lumps of plastic,

它们是大块的塑料，

you know, physical boards that you used to,

就是这些实物板子，

some of you remember, used to actually press the squares down

有些人有印象， 你真的得按下方块，

and there were LED lights, came on.

还会亮起 LED 灯。

And I remember actually, not just thinking about the chess,

我记得我不仅仅是“想着”国际象棋，

I was actually just fascinated by the fact that this lump of plastic,

我是为它着迷，这一块塑料，

someone had programmed it to be smart

有人把它编程得很聪明，

and actually play chess to a really high standard.

以相当高的水平下象棋。

And I was just amazed by that.

我大为震撼。

And that got me thinking about thinking.

这让我开始思考。

And how does the brain come up with these thought processes,

大脑是如何想出这些思维过程、

these ideas,

这些想法的，

and then maybe how we could mimic that with computers.

也许我们该如何用计算机模仿它们。

So yeah, it's been a whole theme for my whole life, really.

所以没错， 这确实是我一生的全部主题。

CA: But you raised all this money to launch DeepMind,

CA：但是你筹集了这些资金 来推出 DeepMind，

and pretty soon you were using it to do, for example, this.

很快你就用它来做一些事， 就比如说这个。

I mean, this is an odd use of it.

这个用法很奇怪。

What was going on here?

怎么回事？

DH: Well, we started off with games at the beginning of DeepMind.

DH：我们从 DeepMind 一开始 就是在玩游戏。

This was back in 2010, so this is from about 10 years ago,

那是 2010 年， 所以是大约 10 年前的事，

it was our first big breakthrough.

这是我们的第一个重大突破。

Because we started off with classic Atari games from the 1970s,

因为我们从 20 世纪 70 年代的 经典雅达利游戏开始，

the simplest kind of computer games there are out there.

是市面上最简单的那种电脑游戏。

And one of the reasons we used games is they're very convenient

我们使用游戏的原因之一 是它们可以非常方便地

to test out your ideas and your algorithms.

测试你的想法和算法。

They're really fast to test.

它们的测试速度非常快。

And also, as your systems get more powerful,

而且，随着你的系统越来越强大，

you can choose harder and harder games.

你可以选择越来越难的游戏。

And this was actually the first time ever that our machine surprised us,

这其实是我们的机器有史以来 第一次让我们感到惊讶，

the first of many times,

许多次中的第一次，

which, it figured out in this game called Breakout,

它在这款名为《打砖块》的游戏中发现，

that you could send the ball round the back of the wall,

你可以让球绕到墙的背面，

and actually, it would be much safer way to knock out all the tiles of the wall.

这样打掉墙上的所有砖块要稳得多。

It's a classic Atari game there.

这是一款经典的雅达利游戏。

And that was our first real aha moment.

那是我们第一个真正的顿悟时刻。

CA: So this thing was not programmed to have any strategy.

CA：所以它并没有 被编程拥有任何策略。

It was just told, try and figure out a way of winning.

只是告诉它， 尝试并找出一个胜出的方法。

You just move the bat at the bottom and see if you can find a way of winning.

只是在底部把击球板移来移去， 看看能不能找到一个胜出的办法。

DH: It was a real revolution at the time.

DH：当时这是一场真正的革命。

So this was in 2012, 2013

2012 年、2013 年，

where we coined these terms "deep reinforcement learning."

我们创造了“深度强化学习”这些术语。

And the key thing about them is that those systems were learning

它们的关键是这些系统

directly from the pixels, the raw pixels on the screen,

直接从像素， 屏幕上的原始像素中学习，

but they weren't being told anything else.

但它们没有被告知其他任何东西。

So they were being told, maximize the score,

它们被要求拿到最高的积分，

here are the pixels on the screen,

这是屏幕上的像素，

30,000 pixels.

30000 个像素。

The system has to make sense on its own from first principles

系统必须根据第一性原理

what’s going on, what it’s controlling,

自行理解正在发生什么、 它要控制什么、

how to get points.

如何获得积分。

And that's the other nice thing about using games to begin with.

这也是从一开始 就使用游戏的另一个好处。

They have clear objectives, to win, to get scores.

它们有明确的目标， 要获胜，要得分。

So you can kind of measure very easily that your systems are improving.

因此，你可以很容易地 衡量你的系统是否在改进。

CA: But there was a direct line from that to this moment

CA：但这直接影响到了 几年后的这个时刻，

a few years later,

where country of South Korea and many other parts of Asia

韩国和亚洲许多其他地区，

and in fact the world went crazy over -- over what?

甚至是全世界都为了什么而沸腾了？

DH: Yeah, so this was the pinnacle of -- this is in 2016 --

DH：是的，这是我们的巅峰， 发生在 2016 年，

the pinnacle of our games-playing work,

我们“玩游戏”成果的巅峰，

where, so we'd done Atari,

我们做了雅达利，

we'd done some more complicated games.

也做了更复杂的游戏。

And then we reached the pinnacle, which was the game of Go,

然后我们达到了巅峰，那就是围棋，

which is what they play in Asia instead of chess,

亚洲人下的是围棋， 而不是国际象棋，

but it's actually more complex than chess.

但它其实比国际象棋更复杂。

And the actual brute force algorithms

用于破解国际象棋的那种暴力算法

that were used to kind of crack chess were not possible with Go

是无法破解围棋的，

because it's a much more pattern-based game,

因为这是一款更加基于模式的游戏，

much more intuitive game.

更直观的游戏。

So even though Deep Blue beat Garry Kasparov in the '90s,

因此，尽管深蓝在 90 年代击败了 加里·卡斯帕罗夫（Garry Kasparov），

it took another 20 years for our program, AlphaGo,

但我们的项目 AlphaGo 又花了 20 年时间

to beat the world champion at Go.

才在围棋比赛中击败世界冠军。

And we always thought,

我们一直认为，

myself and the people working on this project for many years,

我自己和多年从事这个项目的人，

if you could build a system that could beat the world champion at Go,

如果你能构建出一个 能在围棋上击败世界冠军的系统，

it would have had to have done something very interesting.

它一定会做一些非常有趣的事情。

And in this case, what we did with AlphaGo,

在这种情况下， 我们让 AlphaGo 做的

is it basically learned for itself,

基本上就是自主学习，

by playing millions and millions of games against itself,

通过自己和自己玩数百万场游戏

ideas about Go, the right strategies.

学习围棋的思路、正确的策略。

And in fact invented its own new strategies

它实际上研究出了自己的全新策略，

that the Go world had never seen before,

在围棋界都闻所未闻，

even though we've played Go for more than,

即使我们下围棋已有 2000 多年了，

you know, 2,000 years,

it's the oldest board game in existence.

它是现存最古老的棋盘游戏。

So, you know, it was pretty astounding.

这令人瞠目结舌。

Not only did it win the match,

它不仅赢得了比赛，

it also came up with brand new strategies.

还想出了全新的策略。

CA: And you continued this with a new strategy

CA：你继续采用这种新策略，

of not even really teaching it anything about Go,

一点都不教它下围棋，

but just setting up systems

只是建立一个系统，

that just from first principles would play

只凭第一性原则就足以发挥作用，

so that they could teach themselves from scratch, Go or chess.

让它从头自学， 无论是围棋还是国际象棋。

Talk about AlphaZero and the amazing thing that happened in chess then.

谈谈 AlphaZero 以及当时 在国际象棋上发生的神奇故事。

DH: So following this, we started with AlphaGo

DH：在此之后， 我们从 AlphaGo 入手，

by giving it all of the human games that are being played on the internet.

向它提供了网上所有人类玩的游戏。

So it started that as a basic starting point for its knowledge.

它以此为其知识的基本起点。

And then we wanted to see what would happen if we started from scratch,

然后我们想看看如果我们从头开始，

from literally random play.

从随机游戏开始，会发生什么。

So this is what AlphaZero was.

这就是 AlphaZero。

That's why it's the zero in the name,

这就是为什么它名字里是“零”，

because it started with zero prior knowledge

因为它从零先验知识开始。

And the reason we did that is because then we would build a system

而我们之所以这样做， 是因为这样我们就能构建

that was more general.

一个更通用的系统。

So AlphaGo could only play Go,

AlphaGo 只会下围棋，

but AlphaZero could play any two-player game,

但 AlphaZero 会玩所有双人游戏，

and it did it by playing initially randomly

它通过最初随机乱玩，

and then slowly, incrementally improving.

然后缓慢地逐步改进做到这一点。

Well, not very slowly, actually, within the course of 24 hours,

好吧，其实不是很缓慢， 其实是是在 24 小时内

going from random to better than world-champion level.

从乱玩，到更好， 到世界冠军的水平。

CA: And so this is so amazing to me.

CA：这对我来说太神奇了。

So I'm more familiar with chess than with Go.

我对国际象棋比对围棋更熟悉。

And for decades,

几十年来，

thousands and thousands of AI experts worked on building

成千上万的 AI 专家致力于打造

incredible chess computers.

强大的国际象棋计算机。

Eventually, they got better than humans.

最终，它们变得比人类更强。

You had a moment a few years ago,

几年前有过一个时刻，

where in nine hours,

在九个小时内，

AlphaZero taught itself to play chess better than any of those systems ever did.

AlphaZero 自学下国际象棋， 水平超过了任何以往的系统。

Talk about that.

聊聊这个吧。

DH: It was a pretty incredible moment, actually.

DH：那真是一个非常不可思议的时刻。

So we set it going on chess.

我们让它下国际象棋。

And as you said, there's this rich history of chess and AI

正如你所说，国际象棋和 AI 有着悠久的历史，

where there are these expert systems that have been programmed

有用国际象棋思维和算法 编程的专业系统。

with these chess ideas, chess algorithms.

And you have this amazing, you know,

然后就有了这神奇的一刻，

I remember this day very clearly, where you sort of sit down with the system

我清楚记得这一天， 你看着这个系统，

starting off random, you know, in the morning,

从早上开始随便乱下，

you go for a cup of coffee, you come back.

你去喝杯咖啡，然后回来。

I can still just about beat it by lunchtime, maybe just about.

到午餐时间我还能打败它，差不多吧。

And then you let it go for another four hours.

然后让它再下四个小时。

And by dinner,

到了晚饭的时候，

it's the greatest chess-playing entity that's ever existed.

它已经是有史以来 最伟大的国际象棋实体了。

And, you know, it's quite amazing,

这真是太神奇了，

like, looking at that live on something that you know well,

眼睁睁地看着这种，

you know, like chess, and you're expert in

比如国际象棋，你擅长的领域，

and actually just seeing that in front of your eyes.

亲眼目睹它的发生。

And then you extrapolate to what it could then do in science or something else,

然后你推断它在科学 或其他领域也能做什么，

which of course, games were only a means to an end.

当然，游戏只是达到目的的一种手段。

They were never the end in themselves.

它们本身从来不是终点。

They were just the training ground for our ideas

它们只是我们想法的训练场，

and to make quick progress in a matter of, you know,

也是我们快速在一个问题上 取得进展的训练场，

less than five years actually went from Atari to Go.

从雅达利到围棋的时间不到五年。

CA: I mean, this is why people are in awe of AI

CA：这就是为什么 人们对 AI 感到敬畏，

and also kind of terrified by it.

也对它感到恐惧的原因。

I mean, it's not just incremental improvement.

这不仅仅是渐进式的改进。

The fact that in a few hours you can achieve

你可以在几个小时内实现

what millions of humans over centuries have not been able to achieve.

数百万人几个世纪以来 无法实现的目标。

That gives you pause for thought.

这让你停下来思考。

DH: It does, I mean, it's a hugely powerful technology.

DH：确实如此， 它是一项非常强大的技术。

It's going to be incredibly transformative.

这将是极其具有变革性的。

And we have to be very thoughtful about how we use that capability.

我们必须非常仔细地 考虑如何使用这种能力。

CA: So talk about this use of it because this is again,

CA：谈谈它的这种用途， 因为这又是

this is another extension of the work you've done,

你所做工作的又一次延伸，

where now you're turning it to something incredibly useful for the world.

你现正把它变成对世界 非常有用的东西。

What are all the letters on the left, and what’s on the right?

左边的字母都是些什么？ 右边是什么？

DH: This was always my aim with AI from a kid,

DH：这一直是 我小时候使用 AI 的目标，

which is to use it to accelerate scientific discovery.

那就是用它来加速科学发现。

And actually, ever since doing my undergrad at Cambridge,

自从在剑桥大学读本科后，

I had this problem in mind one day for AI,

有一天我想到了 关于 AI 的这个问题，

it's called the protein-folding problem.

叫做蛋白质折叠问题。

And it's kind of like a 50-year grand challenge in biology.

这有点像生物学领域 长达 50 年的重大挑战。

And it's very simple to explain.

而且解释起来很简单。

Proteins are essential to life.

蛋白质对生命至关重要。

They're the building blocks of life.

它们是生命的基石。

Everything in your body depends on proteins.

你体内的一切都依赖于蛋白质。

A protein is sort of described by its amino acid sequence,

蛋白质在某种程度上 是由其氨基酸序列来描述的，

which you can think of as roughly the genetic sequence

你可以把它大致看作是 描述蛋白质的基因序列，

describing the protein, so that are the letters.

这些字母就是这个序列。

CA: And each of those letters represents in itself a complex molecule?

CA：每一个字母 都代表一个复杂的分子？

DH: That's right, each of those letters is an amino acid.

DH：没错，每个字母都是一个氨基酸。

And you can think of them as a kind of string of beads

你可以把它们看成一串串珠子，

there at the bottom, left, right?

出现在底部、左边、右边，对吧？

But in nature, in your body or in an animal,

但是在大自然中， 在你或者动物的身体里，

this string, a sequence,

这一串字母，这个序列，

turns into this beautiful shape on the right.

会变成右边这个漂亮的形状。

That's the protein.

这就是蛋白质。

Those letters describe that shape.

这些字母描述了这个形状。

And that's what it looks like in nature.

这就是它在自然界中的样子。

And the important thing about that 3D structure is

而这个三维结构的重要之处在于

the 3D structure of the protein goes a long way to telling you

蛋白质的三维结构 在很大程度上可以告诉你

what its function is in the body, what it does.

它在体内的功能、它的作用。

And so the protein-folding problem is:

蛋白质折叠问题是：

Can you directly predict the 3D structure just from the amino acid sequence?

你能直接从氨基酸序列中 预测三维结构吗？

So literally if you give the machine, the AI system,

也就是如果你给机器、AI 系统

the letters on the left,

左边的字母，

can it produce the 3D structure on the right?

它能生成右边的三维结构吗？

And that's what AlphaFold does, our program does.

AlphaFold 做的就是这件事， 我们的程序做的就是这件事。

CA: It's not calculating it from the letters,

CA：它不是根据字母计算出来的，

it's looking at patterns of other folded proteins that are known about

而是研究已知的 其他折叠蛋白质的模式，

and somehow learning from those patterns

以某种方式从这些模式中学到

that this may be the way to do this?

可能做到这一点的方法？

DH: So when we started this project, actually straight after AlphaGo,

DH：我们开始做这个项目时， 其实是紧接着 AlphaGo，

I thought we were ready.

我以为我们已经准备好了。

Once we'd cracked Go,

当我们破解了围棋之后，

I felt we were finally ready after, you know,

我以为我们终于 在研究了 20 年之后准备就绪，

almost 20 years of working on this stuff

to actually tackle some scientific problems,

可以真正解决一些科学问题，

including protein folding.

包括蛋白质折叠。

And what we start with is painstakingly,

我们一开始就是勤勤恳恳地，

over the last 40-plus years,

在过去的 40 多年里，

experimental biologists have pieced together

实验生物学家拼出了

around 150,000 protein structures

大约 15 万个蛋白质结构，

using very complicated, you know, X-ray crystallography techniques

借助非常复杂的 X 射线晶体学技术

and other complicated experimental techniques.

和其他复杂的实验技术。

And the rule of thumb is

粗略估算，

that it takes one PhD student their whole PhD,

一个博士生需要 花费整个博士学位期间，

so four or five years, to uncover one structure.

也就是四到五年的时间， 才能解析一个结构。

But there are 200 million proteins known to nature.

但是大自然 已知有 2 亿种蛋白质。

So you could just, you know, take forever to do that.

你可以花上无限的时间。

And so we managed to actually fold, using AlphaFold, in one year,

我们使用 AlphaFold 在一年内折叠了

all those 200 million proteins known to science.

科学已知的所有 2 亿种蛋白质。

So that's a billion years of PhD time saved.

给博士生省了十亿年。

(Applause)

（掌声）

CA: So it's amazing to me just how reliably it works.

CA：它的可靠性令我惊艳。

I mean, this shows, you know,

这表明，这就是模型， 你做了实验。

here's the model and you do the experiment.

And sure enough, the protein turns out the same way.

果然，蛋白质就是这样的。

Times 200 million.

乘以 2 亿。

DH: And the more deeply you go into proteins,

DH：你对蛋白质的了解越深，

you just start appreciating how exquisite they are.

你就会体会到它们有多精巧。

I mean, look at how beautiful these proteins are.

看看这些蛋白质多漂亮啊。

And each of these things do a special function in nature.

这些东西在自然界中 都起着特殊的作用。

And they're almost like works of art.

它们几乎就像是艺术品。

And it's still astounds me today that AlphaFold can predict,

如今依然令我震惊的是， AlphaFold 能够预测，

the green is the ground truth, and the blue is the prediction,

绿色是基本事实， 蓝色是预测，

how well it can predict, is to within the width of an atom on average,

它能预测得多好， 平均在一个原子的宽度以内，

is how accurate the prediction is,

它的预测就是这么精确，

which is what is needed for biologists to use it,

生物学家要使用它 就需要这样的精确度，

and for drug design and for disease understanding,

也是设计药物和了解疾病所需要的，

which is what AlphaFold unlocks.

而这正是 AlphaFold 所能带来的。

CA: You made a surprising decision,

CA：你做了一个出人意料的决定，

which was to give away the actual results of your 200 million proteins.

那就是将 2 亿种蛋白质的实际结果 “拱手相让”。

DH: We open-sourced AlphaFold and gave everything away

DH：我们开源了 AlphaFold，并把一切

on a huge database with our wonderful colleagues,

都通过一个巨大的数据库 交给了我们优秀的同行，

the European Bioinformatics Institute.

他们来自欧洲生物信息学研究所。

(Applause)

（掌声）

CA: I mean, you're part of Google.

CA：你是谷歌的一员。

Was there a phone call saying, "Uh, Demis, what did you just do?"

有没有人打电话说： “呃，戴密斯，你刚才做了什么？”

DH: You know, I'm lucky we have very supportive,

DH：我很幸运 我们得到了非常大的支持，

Google's really supportive of science

谷歌非常支持科学，

and understand the benefits this can bring to the world.

也知道这能给世界带来什么好处。

And, you know, the argument here

而且问题是

was that we could only ever have even scratched the surface

我们做到的只是触及

of the potential of what we could do with this.

我们能用它做些什么的潜力的表面。

This, you know, maybe like a millionth

这大概是科学界正在 用它做的事情的百万分之一。

of what the scientific community is doing with it.

There's over a million and a half biologists around the world

全世界有超过 150 万名生物学家

have used AlphaFold and its predictions.

使用了 AlphaFold 及其预测。

We think that's almost every biologist in the world

我们认为，世界上几乎每个生物学家、

is making use of this now, every pharma company.

每家制药公司都在使用它。

So we'll never know probably what the full impact of it all is.

因此，我们可能永远不会知道 这一切的全部影响。

CA: But you're continuing this work in a new company

CA：但是你正在从谷歌拆分出来的

that's spinning out of Google called Isomorph.

一家名为 Isomorph 的新公司里 继续这项工作。

DH: Isomorphic.

DH：Isomorphic。

CA: Isomorphic.

CA：Isomorphic。

Give us just a sense of the vision there.

让我们了解一下它的愿景。 愿景是什么？

What's the vision?

DH：AlphaFold 是一种基础生物学工具。

DH: AlphaFold is a sort of fundamental biology tool.

Like, what are these 3D structures,

比如，这些三维结构是什么，

and then what might they do in nature?

它们在自然界中会做些什么？

And then if you, you know,

如果你……我之所以想到这个问题 并对此感到非常兴奋，

the reason I thought about this and was so excited about this,

is that this is the beginnings of understanding disease

是因为这是了解疾病的开始，

and also maybe helpful for designing drugs.

也可能对药物的设计有所帮助。

So if you know the shape of the protein,

如果你知道蛋白质的形状，

and then you can kind of figure out

你就能弄清楚蛋白质表面的哪一部分

which part of the surface of the protein

you're going to target with your drug compound.

是你的药物化合物的靶点。

And Isomorphic is extending this work we did in AlphaFold

Isomorphic 正在将我们 在 AlphaFold 中做的这项工作

into the chemistry space,

扩展到药物领域，

where we can design chemical compounds

我们可以设计出能够精确 与蛋白质正确位置结合的化合物，

that will bind exactly to the right spot on the protein

and also, importantly, to nothing else in the body.

更重要的是， 它不会与体内其他任何东西结合。

So it doesn't have any side effects and it's not toxic and so on.

所以它没有任何副作用， 也没有毒性等等。

And we're building many other AI models,

而且我们正在构建 许多其他 AI 模型，

sort of sister models to AlphaFold

类似于 AlphaFold 的姊妹模型帮助预测，

to help predict,

make predictions in chemistry space.

在药物领域做出预测。

CA: So we can expect to see

CA：我们可以预见，

some pretty dramatic health medicine breakthroughs

在未来几年中，健康医学 将取得一些相当显著的突破。

in the coming few years.

DH: I think we'll be able to get down drug discovery

DH：我认为我们可以 将药物研发时间

from years to maybe months.

从几年缩短到几个月。

CA: OK. Demis, I'd like to change direction a bit.

CA：好吧。戴密斯， 我想稍微转换一下话题。

Our mutual friend, Liv Boeree, gave a talk last year at TEDAI

我们共同的朋友 丽芙·波瑞（Liv Boeree）

去年在 TEDAI 上发表了一场演讲，

that she called the “Moloch Trap.”

她称之为《摩洛克陷阱》 （Moloch Trap）。

The Moloch Trap is a situation

摩洛克陷阱指的是

where organizations,

处于竞争环境中的组织

companies in a competitive situation can be driven to do things

和公司可能被迫去做

that no individual running those companies would by themselves do.

任何一个经营这些公司的人 自己不会做的事情。

I was really struck by this talk,

这场演讲真的给我留下了深刻的印象，

and it's felt, as a sort of layperson observer,

作为一个外行观察者，

that the Moloch Trap has been shockingly in effect in the last couple of years.

我觉得在过去的几年里， 摩洛克陷阱的作用令人震惊。

So here you are with DeepMind,

你在 DeepMind，

sort of pursuing these amazing medical breakthroughs

在追求这些惊人的医学突破

and scientific breakthroughs,

和科学突破，

and then suddenly, kind of out of left field,

然后突然之间，

微软投资的 OpenAI发布了 ChatGPT， 遥遥领先。

OpenAI with Microsoft releases ChatGPT.

And the world goes crazy and suddenly goes, “Holy crap, AI is ...”

然后世界都为之疯狂， 突然就变成：“天哪，AI 是……”，

you know, everyone can use it.

每个人都可以使用它。

And there’s a sort of, it felt like the Moloch Trap in action.

就像是亲身实践摩洛克陷阱。

I think Microsoft CEO Satya Nadella actually said,

我记得微软首席执行官 萨提亚·纳德拉（Satya Nadella）说过：

"Google is the 800-pound gorilla in the search space.

“谷歌是搜索领域的霸主。

We wanted to make Google dance."

我们想让谷歌手忙脚乱。”

How ...?

怎么……？

And it did, Google did dance.

确实如此，谷歌确实手忙脚乱。

There was a dramatic response.

引起了剧烈的反应。

Your role was changed,

你的角色发生了变化，

you took over the whole Google AI effort.

你接管了谷歌 AI 的全部工作。

Products were rushed out.

产品匆匆上线。

You know, Gemini, some part amazing, part embarrassing.

Gemini 有一部分很神奇， 有一部分很丢人。

I’m not going to ask you about Gemini because you’ve addressed it elsewhere.

我不会问你有关 Gemini 的问题， 因为你在别处已经解答过了。

But it feels like this was the Moloch Trap happening,

但这感觉就像是摩洛克陷阱正在上演，

that you and others were pushed to do stuff

你和其他人被迫去做

that you wouldn't have done without this sort of catalyzing competitive thing.

没有竞争刺激你根本不会做的事情。

Meta did something similar as well.

Meta 也做了类似的事情。

They rushed out an open-source version of AI,

他们匆忙推出了 AI 的开源版本，

which is arguably a reckless act in itself.

这本身可以说是一种鲁莽的行为。

This seems terrifying to me.

这对我来说很可怕。

Is it terrifying?

可怕吗？

DH: Look, it's a complicated topic, of course.

DH：这当然是一个复杂的话题。

And, first of all, I mean, there are many things to say about it.

首先，这个话题有很多可说的。

First of all, we were working on many large language models.

首先，我们正在研究 很多大语言模型。

And in fact, obviously, Google research actually invented Transformers,

很明显，谷歌研究在五、六年前 发明了 Transformer 模型，

as you know,

which was the architecture that allowed all this to be possible,

正是这种架构使这一切成为可能。

five, six years ago.

And so we had many large models internally.

我们内部有许多大模型。

The thing was, I think what the ChatGPT moment did that changed was,

问题是，我认为 ChatGPT 时刻 所带来的改变是，

and fair play to them to do that, was they demonstrated,

他们这么做也是公平的， 因为他们证明了，

I think somewhat surprisingly to themselves as well,

我觉得对他们自己来说也是意外的，

that the public were ready to,

他们证明了公众已经准备好，

you know, the general public were ready to embrace these systems

公众已经准备好接受这些系统

and actually find value in these systems.

并从中寻找价值。

Impressive though they are, I guess, when we're working on these systems,

虽然它们很厉害， 但是我们在研究这些系统的时候，

mostly you're focusing on the flaws and the things they don't do

主要关注的是缺陷 和它们做不了的事情，

and hallucinations and things you're all familiar with now.

还有幻觉和你们现在都熟知的问题。

We're thinking, you know,

我们在想，

would anyone really find that useful given that it does this and that?

它有这种问题，有那种问题， 真的会有人觉得它有用吗？

And we would want them to improve those things first,

我们希望它们能在推出之前 先改进这些问题。

before putting them out.

But interestingly, it turned out that even with those flaws,

但有趣的是，事实证明， 即使有这些缺陷，

many tens of millions of people still find them very useful.

仍有数千万人认为它们非常有用。

And so that was an interesting update on maybe the convergence of products

这是产品和科学融合的

and the science that actually,

一次有趣的进步，

all of these amazing things we've been doing in the lab, so to speak,

我们在实验室里做的 各种有意义的事，

are actually ready for prime time for general use,

都已经准备好 登上通用的大舞台，

beyond the rarefied world of science.

走出科学世界的象牙塔。

And I think that's pretty exciting in many ways.

我认为这在很多方面 都非常令人兴奋。

CA: So at the moment, we've got this exciting array of products

CA：目前，我们有一系列 令人兴奋的产品，

which we're all enjoying.

我们都很喜欢。

And, you know, all this generative AI stuff is amazing.

所有这些生成式 AI 的东西 都太神奇了。

But let's roll the clock forward a bit.

但让我们把时间向前推一点。

Microsoft and OpenAI are reported to be building

据报道，微软和 OpenAI 正在打造

or investing like 100 billion dollars

或投资约 1000 亿美元，

into an absolute monster database supercomputer

建造一台绝对庞大的 数据库超级计算机，

that can offer compute at orders of magnitude

它提供的计算能力比我们当今的 任何东西都要高出几个数量级。

more than anything we have today.

It takes like five gigawatts of energy to drive this, it's estimated.

据估计，驱动它 大约需要五千兆瓦的能量。

That's the energy of New York City to drive a data center.

驱动一个数据中心 要花上整个纽约市的能量。

So we're pumping all this energy into this giant, vast brain.

我们将这些能量 注入这个巨型、庞大的大脑中。

Google, I presume is going to match this type of investment, right?

我相信谷歌也会进行相应的投资，对吧？

DH: Well, I mean, we don't talk about our specific numbers,

DH：我们不谈论我们的具体数字，

but you know, I think we're investing more than that over time.

但我认为我们长期以来的投资 不止这个数。

So, and that's one of the reasons

因此，这也是我们 在 2014 年与谷歌合作的原因之一，

we teamed up with Google back in 2014,

is kind of we knew that in order to get to AGI,

因为我们知道要实现 AGI，

we would need a lot of compute.

我们需要大量的计算。

And that's what's transpired.

众所周知。

And Google, you know, had and still has the most computers.

谷歌无论过去还是现在 都有着最多的计算机。

CA: So Earth is building these giant computers

CA：地球正在打造 这些巨型计算机，

that are going to basically, these giant brains,

这些巨型大脑，

that are going to power so much of the future economy.

将大大助力未来的经济。

And it's all by companies that are in competition with each other.

而这一切都是公司互相竞争带来的。

How will we avoid the situation where someone is getting a lead,

我们将如何避免 有人一拿到了消息，

someone else has got 100 billion dollars invested in their thing.

另一波人就向自己这儿 投了 1000 亿美元。

Isn't someone going to go, "Wait a sec.

难道没有人会说：“等一下。

If we used reinforcement learning here

如果我们在这里使用强化学习

to maybe have the AI tweak its own code

让 AI 调整自己的代码，

and rewrite itself and make it so [powerful],

重写自己的代码， 让它变得[强大]，

we might be able to catch up in nine hours over the weekend

我们也许可以在周末 花上九个小时就能

with what they're doing.

做到他们现在正在做的事。

Roll the dice, dammit, we have no choice.

赌一把吧，该死， 我们别无选择。

Otherwise we're going to lose a fortune for our shareholders."

不然我们就要 让我们的股东亏一大笔钱了。”

How are we going to avoid that?

我们要如何避免这种情况？

DH: Yeah, well, we must avoid that, of course, clearly.

DH：是的，当然， 我们必须要避免这种情况。

And my view is that as we get closer to AGI,

我的观点是， 随着我们距离 AGI 越来越近，

we need to collaborate more.

我们需要加强合作。

And the good news is that most of the scientists involved in these labs

好消息是，参与这些实验室的 大多数科学家

know each other very well.

彼此非常了解。

And we talk to each other a lot at conferences and other things.

我们经常在会议和其他场合交流。

And this technology is still relatively nascent.

这项技术还相对处于起步阶段。

So probably it's OK what's happening at the moment.

所以目前发生的事情可能没问题。

But as we get closer to AGI, I think as a society,

但是随着我们越来越接近 AGI， 我认为从社会整体来看，

we need to start thinking about the types of architectures that get built.

我们需要开始考虑 我们要构建的架构。

So I'm very optimistic, of course,

当然，我非常乐观，

that's why I spent my whole life working on AI and working towards AGI.

这就是为什么我一生都在研究 AI 并努力实现 AGI 的原因。

But I suspect there are many ways to build the architecture safely, robustly,

但我怀疑有很多方法 可以以安全、稳健、

reliably and in an understandable way.

可靠且易于理解的方式构建架构。

And I think there are almost certainly going to be ways of building architectures

而且我认为几乎肯定 存在或多或少不安全或有风险的

that are unsafe or risky in some form.

构建架构的方法。

So I see a sort of,

我看到了一种， 我们必须让人类跨过的瓶颈，

a kind of bottleneck that we have to get humanity through,

which is building safe architectures as the first types of AGI systems.

就是将安全的架构 打造成最初始的 AGI 系统。

And then after that, we can have a sort of,

在那之后， 我们可以

a flourishing of many different types of systems

让许多不同类型的系统蓬勃发展，

that are perhaps sharded off those safe architectures

可能与那些在理想情况下

that ideally have some mathematical guarantees

有一些数学保证

or at least some practical guarantees around what they do.

或至少有一些实际保证的 安全架构分开。

CA: Do governments have an essential role here

CA：在定义公平竞争环境 和绝对禁忌方面，

to define what a level playing field looks like

政府在这方面 是否起着至关重要的作用？

and what is absolutely taboo?

DH: Yeah, I think it's not just about --

DH：是的，我认为这不仅仅是——

actually I think government and civil society

我认为政府、民间社会、

and academia and all parts of society have a critical role to play here

学术界和社会各部分 都可以在这里发挥至关重要的作用，

to shape, along with industry labs,

与行业实验室一起，

what that should look like as we get closer to AGI

塑造当我们更接近 AGI 时， 它们该有的样子，

and the cooperation needed and the collaboration needed,

需要企业，需要合作，

to prevent that kind of runaway race dynamic happening.

防止这种失控竞赛局面的出现。

CA: OK, well, it sounds like you remain optimistic.

CA：听起来你还是很乐观的。

What's this image here?

这张照片是什么？

DH: That's one of my favorite images, actually.

DH：这是我最喜欢的照片之一。

I call it, like, the tree of all knowledge.

我把它叫做全知之树。

So, you know, we've been talking a lot about science,

我们一直在谈论科学，

and a lot of science can be boiled down to

很多科学可以归结为……

if you imagine all the knowledge that exists in the world

如果你将世界上存在的所有知识

as a tree of knowledge,

想象成一棵知识树，

and then maybe what we know today as a civilization is some, you know,

那么我们今天所知道的文明可能只是

small subset of that.

其中的一小部分。

And I see AI as this tool that allows us,

我认为 AI 一种工具，它使我们

as scientists, to explore, potentially, the entire tree one day.

作为科学家 有朝一日能够探索整棵树。

And we have this idea of root node problems

我们有“根节点”问题，

that, like AlphaFold, the protein-folding problem,

比如 AlphaFold， 蛋白质折叠问题，

where if you could crack them,

如果你能破解它，

it unlocks an entire new branch of discovery or new research.

它就会为发现或新研究 开辟出一个全新的分支。

And that's what we try and focus on at DeepMind

这就是我们在 DeepMind

and Google DeepMind to crack those.

和谷歌 DeepMind 努力征服 和关注的点。

And if we get this right, then I think we could be, you know,

如果我们做对了，那么我认为，

in this incredible new era of radical abundance,

我们就可以进入 这个相当富足的美妙新时代，

curing all diseases,

治愈所有疾病，

spreading consciousness to the stars.

向群星传播意识。

You know, maximum human flourishing.

人类的极度繁荣。

CA: We're out of time,

CA：我们没时间了，

but what's the last example of like, in your dreams,

但是最后举个例子， 在你的梦想中，

this dream question that you think there is a shot

这个梦中的问题是： 你认为在你有生之年，

that in your lifetime AI might take us to?

AI 能带我们去向何方？

DH: I mean, once AGI is built,

DH：一旦建成 AGI，

what I'd like to use it for is to try and use it to understand

我想用它理解现实的基本本质。

the fundamental nature of reality.

So do experiments at the Planck scale.

在普朗克尺度的极限做实验。

You know, the smallest possible scale, theoretical scale,

极限小的尺度，理论尺度，

which is almost like the resolution of reality.

就像是现实的分辨率。

CA: You know, I was brought up religious.

CA：我从小就信奉宗教。

And in the Bible, there’s a story about the tree of knowledge

在《圣经》中，有一个 关于知识之树的故事，

that doesn't work out very well.

但结果不怎么样。

(Laughter)

（笑声）

Is there any scenario

有没有什么情形

where we discover knowledge that the universe says,

会让我们发现这样的知识，宇宙说：

"Humans, you may not know that."

“人类，你可能还不知道吧。”

DH: Potentially.

DH：有可能。

I mean, there might be some unknowable things.

可能有一些不得而知的事情。

But I think scientific method is the greatest sort of invention

但我认为，科学方法是

人类有史以来最伟大的发明。

humans have ever come up with.

You know, the enlightenment and scientific discovery.

启蒙运动和科学发现。

That's what's built this incredible modern civilization around us

它们铸就了我们身处的伟大现代文明

and all the tools that we use.

和我们使用的各种工具。

So I think it's the best technique we have

我认为这是我们 了解身边浩渺宇宙的最佳方式。

for understanding the enormity of the universe around us.

CA: Well, Demis, you've already changed the world.

CA：戴密斯，你已经改变了世界。

I think probably everyone here will be cheering you on

我想可能在座的每个人 都会为你加油，

in your efforts to ensure that we continue to accelerate

努力争取我们继续 朝着正确的方向加速。

in the right direction.

DH: Thank you.

DH：谢谢。

CA: Demis Hassabis.

CA：戴密斯·哈萨比斯。

(Applause)

（掌声）